Panoptic Scene Graph Generation - Technology Org

Scene graph generation generates a graph-structured representation from a given image to abstract out objects--grounded by bounding boxes--and their pairwise relationships. It has diverse applications, such as visual reasoning, image captioning, and robotics. They only provide a coarse localization of objects and cannot cover the full scene of an image. Researchers propose panoptic scene graph generation (PSG) for generating scene graph representations based on panoptic segmentations rather than rigid bounding boxes. A large PSG dataset with high-quality annotations is created, and two-stage and one-stage PSG baselines are proposed.